Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 9: French-German Session: Simulation Methods and Modeling of Soft Matter I
CPP 9.4: Talk
Monday, March 9, 2026, 16:00–16:15, ZEU/LICH
Machine learning potentials for redox chemistry in solution — •Redouan El Haouari1,2, Emir Kocer1,2, and Jörg Behler1,2 — 1Theoretische Chemie II, Ruhr-Universität Bochum, Germany — 2Research Center Chemical Sciences and Sustainability, Research Alliance Ruhr, Germany
Most Machine-Learning Potentials (MLPs) currently in use are local, which prevents the construction of accurate potentials for systems in which long-range charge transfer is important. One example is the distinction of different oxidation states of transition metal ions such as ferrous (Fe2+) and ferric (Fe3+) ions in aqueous solution. In this case the near-sightedness of local MLPs cannot account for distant counter ions or non-local charge transfer. We show that 4th-Generation High-Dimensional Neural Network Potentials (4G-HDNNPs), which employ atomic charges obtained from charge equilibration as globally-determined descriptors, do not suffer from this short-coming. For aqueous ferric and ferrous chloride, the model predicts iron oxidation states correctly matching the total number of chloride ions in the system demonstrating that physical knowledge about the system of interest remains essential to construct reliable MLPs.
Keywords: Machine-Learning Potentials; High-Dimensional Neural Network Potentials; 4G-HDNNPs; Charge Transfer; Redox Chemistry in Solution